Master Motor Map
Master Motor Map (MMM) is a conceptual framework for perception, visualization, reproduction, and recognition of human motion in order to decouple motion capture data from further post-processing tasks, such as execution on a real humanoid robot. Employing MMM makes it easy to map motions between different kinematics independently and uniformly as well as to analyze certain dynamic aspects of the considered motion.
Part of this framework is a dynamic model consisting of a particular kinematic structure enriched with pre-defined segment properties (anthropometric data) e.g. mass distribution, segment length, moment of inertia, etc. Moreover, the strategy is to define the maximum number of DoFs that might be used by any visualization, recognition, or reproduction module.
Rigid Body Dynamics Library
RBDL is a highly efficient C++ library that contains some essential rigid body dynamics algorithms such as the Articulated Body Algorithm (ABA) for forward dynamics, Recursive Newton-Euler Algorithm (RNEA) for inverse dynamics and the Composite Rigid Body Algorithm (CRBA) for the efficient computation of the joint space inertia matrix. It further contains code for Jacobians, forward and inverse kinematics, and handling of external constraints such as contacts and collisions.
The code is developed by Martin Felis at the research group Optimization in Robotics and Biomechanics (ORB) of the Interdisciplinary Center for Scientific Computing (IWR) at Heidelberg University. The code tightly follows the notation used in Roy Featherstone”s book “Rigid Body Dynamics Algorithm”.
Generic Reinforcement Learning Library
GRL is a Generic Reinforcement Learning Library developed within the Koroibot project. The library allows users to define an experiment (of online or batch type), which is comprised of one environment and one or two agents (controllers). The environment is defined by a dynamical model of a system (like one of the Koroibot bipedal robots) and a task to execute (like walking with a reference velocity). The agents can be of multiple types, such as fixed (PID, NMPC), or learning (RL), and there can be switching or sequential execution of these two agents. The library includes powerful interfaces, which allow easy and fast (re-)configuration of environments, agents, RL learning algorithms and representations of functions or visualization methods.
The library contains several examples, particularly a model of the bipedal robot LEO used in Koroibot, and an interface to MUSCOD II, to enable combination of RL and NMPC.